有姿态变化的人脸图像识别方法论文

编号：审定成绩：重庆邮电大学毕业设计（论文）设计（论文）题目：有姿态变化的人脸图像识别方法学院名称：专业：班级：学号：指导教师：答辩组负责人：填表时间：2014年6月重庆邮电大学教务处制摘要人脸识别是一种研究以生物特征为基础的使用计算机实现的识别技术。在生物特征识别领域，人脸识别技术操作简单快速，结果直观，准确可靠，不需要人为的配合，正是这些优点使其成为一个热门学科。人脸识别在人工智能领域中占有极其重要的地位，是由于它自身的优越性以及在计算机视觉、图像处理、模式识别、多媒体、心理学等多个领域的广泛运用。本文研究的重点正是在姿态变化下的多姿态人脸识别，研究测试样本出现姿态变化时的人脸识别的问题。在实际应用中，我们常常只能获得极少量的训练样本。然而传统的方法在处理具有姿态变化的测试样本时，识别率会急速下降，从而导致系统无法正常工作。为了研究数据更易处理、快速、较易在实际应用中推广的人脸识别算法和方案，本文主要讨论了专门针对有姿态变化姿态样本的人脸识别策略。本文按照完整人脸识别流程来分析基于PCA(主成分分析)的人脸识别算法实现的性能,选用了ORL人脸数据库。从人脸识别的实效性和可操作性两个角度考虑，针对平面内具有随机旋转角度的人脸图像难以识别问题，本文研究了基于二维图像的正弦变换方法，其特点是可以实现快速的姿态校正。本文使用PCA对训练集进行人脸特征提取，使用KNN（K近邻）分类器来进行人脸判别分类。最后比较有无姿态校正模块的人脸识别系统的准确率。【关键词】人脸识别姿态校正正弦变换PCA算法K最近邻分类器ABSTRACTFacerecognitionisacomputerrecognitiontechnology,whichisbasedonbiologicalcharacteristics.Inthefieldofbiometricidentification,theadvantagesoffacerecognitionarefastandsimpleandtheresultsareintuitive,accurateandreliable.Ithasbecomeapopularsubjectforitdoesnotneedco-ordination.Duetoitsadvantages,facerecognitionhasbeenwidelyusedincomputervision,imageprocessing,patternrecognition,multimediaprocessing,psychologyandotherfields.Soitplaysanimportantroleinthefieldofartificialintelligence.Thisstudyistofocusonpose-variedfacerecognitionwiththechangeofpostureandresearchthefacerecognitionproblemwhentheposturesoftestingsampleshavechanged.Inpractice,wemerelygetaminimalamountoftrainingsamples.Iftestsampleshaveencounteredtheposturechanging,therecognitionrateofthetraditionalmethodswouldreducedramatically,whichcausesthesystemtobeoutofwork.Thispaperfocusesspecificallyondevelopingfacerecognitionstrategyofmulti-gesturesamples,whichmakesdataeasytohandle,andpromotingthepracticalapplicationoffacerecognitionalgorithmsandprograms.ThispaperanalyzestheperformanceoffacerecognitionalgorithmwhichbasedonthePCA(PrincipalComponentAnalysis)andtheexperimentsareconductedontheORLfacedatabase.Consideringtheeffectivenessandoperabilityoffacerecognitionandthedifficultiesofidentifyingthefaceimageswithrandomrotationangle,thispaperstudiesthesinusoidaltransformbasedontwo-dimensionalimage,whichcanachieverapidposturecorrection.TrainingsetusingPCAtoextractfacialfeatures,andusingKNN(KNearestNeighbor)forfaceidentificationandclassificationinthispaper.Finally,comparingtheaccuracyoffacerecognitionsystemwhichhasposturecorrectionmoduleornot.【Keywords】facerecognitionposecorrectedsinusoidaltransformPCAalgorithmKNN目录前言.............................................................................................................................................-1-第一章绪论.................................................................................................................................-2-第一节研究背景及意义.....................................................................................................-2-第二节人脸识别研究现状.................................................................................................-3-一、人脸识别现状.........................................................................................................-3-二、多姿态人脸识别现状.............................................................................................-5-第三节人脸识别的应用前景.............................................................................................-7-第四节本章小结.................................................................................................................-9-第二章人脸识别技术概述.......................................................................................................-10-第一节人脸识别的研究范围...........................................................................................-10-第二节人脸识别系统.......................................................................................................-10-第三节人脸识别的性能评价...........................................................................................-11-第四节本章小结...............................................................................................................-12-第三章多姿态人脸识别的研究...............................................................................................-13-第一节引言.......................................................................................................................-13-第二节多姿态人脸识别策略...........................................................................................-13-第三节姿态校正研究.......................................................................................................-14-一、姿态校正的人脸识别框架...................................................................................-14-二、基于ST的姿态校正算法....................................................................................-15-三、实验结果...............................................................................................................-15-第四节本章小结...............................................................................................................-19-第四章基于PCA的人脸识别方法.........................................................................................-20-第一节问题描述................................................................................................